Don't copy fill array data to quick literal pool.
Currently quick copies the fill array data from the dex file to the literal
pool. It then has to go through hoops to pass this PC relative address down
to out-of-line code. Instead, pass the offset of the table to the out-of-line
code and use the CodeItem data associated with the ArtMethod. This reduces
the size of oat code while greatly simplifying it.
Unify the FillArrayData implementation in quick, portable and the interpreters.
Change-Id: I9c6971cf46285fbf197856627368c0185fdc98ca
diff --git a/compiler/dex/mir_graph.h b/compiler/dex/mir_graph.h
index f14b187..fe6fb75 100644
--- a/compiler/dex/mir_graph.h
+++ b/compiler/dex/mir_graph.h
@@ -338,7 +338,7 @@
uint32_t method_lowering_info;
} meta;
- explicit MIR():offset(0), optimization_flags(0), m_unit_index(0), bb(NullBasicBlockId),
+ explicit MIR() : offset(0), optimization_flags(0), m_unit_index(0), bb(NullBasicBlockId),
next(nullptr), ssa_rep(nullptr) {
memset(&meta, 0, sizeof(meta));
}
@@ -572,7 +572,7 @@
* @return Returns the raw table pointer.
*/
const uint16_t* GetTable(MIR* mir, uint32_t table_offset) const {
- return GetInsns(mir->m_unit_index) + mir->offset + table_offset;
+ return GetInsns(mir->m_unit_index) + mir->offset + static_cast<int32_t>(table_offset);
}
unsigned int GetNumBlocks() const {
diff --git a/compiler/dex/quick/arm/call_arm.cc b/compiler/dex/quick/arm/call_arm.cc
index b721e02..13b9bf0 100644
--- a/compiler/dex/quick/arm/call_arm.cc
+++ b/compiler/dex/quick/arm/call_arm.cc
@@ -140,41 +140,6 @@
}
/*
- * Array data table format:
- * ushort ident = 0x0300 magic value
- * ushort width width of each element in the table
- * uint size number of elements in the table
- * ubyte data[size*width] table of data values (may contain a single-byte
- * padding at the end)
- *
- * Total size is 4+(width * size + 1)/2 16-bit code units.
- */
-void ArmMir2Lir::GenFillArrayData(MIR* mir, DexOffset table_offset, RegLocation rl_src) {
- const uint16_t* table = mir_graph_->GetTable(mir, table_offset);
- // Add the table to the list - we'll process it later
- FillArrayData *tab_rec =
- static_cast<FillArrayData*>(arena_->Alloc(sizeof(FillArrayData), kArenaAllocData));
- tab_rec->table = table;
- tab_rec->vaddr = current_dalvik_offset_;
- uint16_t width = tab_rec->table[1];
- uint32_t size = tab_rec->table[2] | ((static_cast<uint32_t>(tab_rec->table[3])) << 16);
- tab_rec->size = (size * width) + 8;
-
- fill_array_data_.push_back(tab_rec);
-
- // Making a call - use explicit registers
- FlushAllRegs(); /* Everything to home location */
- LoadValueDirectFixed(rl_src, rs_r0);
- LoadWordDisp(rs_rARM_SELF, QUICK_ENTRYPOINT_OFFSET(4, pHandleFillArrayData).Int32Value(),
- rs_rARM_LR);
- // Materialize a pointer to the fill data image
- NewLIR3(kThumb2Adr, rs_r1.GetReg(), 0, WrapPointer(tab_rec));
- ClobberCallerSave();
- LIR* call_inst = OpReg(kOpBlx, rs_rARM_LR);
- MarkSafepointPC(call_inst);
-}
-
-/*
* Handle unlocked -> thin locked transition inline or else call out to quick entrypoint. For more
* details see monitor.cc.
*/
diff --git a/compiler/dex/quick/arm/codegen_arm.h b/compiler/dex/quick/arm/codegen_arm.h
index 932dd87..1c87a03 100644
--- a/compiler/dex/quick/arm/codegen_arm.h
+++ b/compiler/dex/quick/arm/codegen_arm.h
@@ -118,7 +118,6 @@
void GenEntrySequence(RegLocation* ArgLocs, RegLocation rl_method);
void GenExitSequence();
void GenSpecialExitSequence();
- void GenFillArrayData(MIR* mir, DexOffset table_offset, RegLocation rl_src);
void GenFusedFPCmpBranch(BasicBlock* bb, MIR* mir, bool gt_bias, bool is_double);
void GenFusedLongCmpBranch(BasicBlock* bb, MIR* mir);
void GenSelect(BasicBlock* bb, MIR* mir);
diff --git a/compiler/dex/quick/arm64/call_arm64.cc b/compiler/dex/quick/arm64/call_arm64.cc
index e8de876..c898e2d 100644
--- a/compiler/dex/quick/arm64/call_arm64.cc
+++ b/compiler/dex/quick/arm64/call_arm64.cc
@@ -149,41 +149,6 @@
}
/*
- * Array data table format:
- * ushort ident = 0x0300 magic value
- * ushort width width of each element in the table
- * uint size number of elements in the table
- * ubyte data[size*width] table of data values (may contain a single-byte
- * padding at the end)
- *
- * Total size is 4+(width * size + 1)/2 16-bit code units.
- */
-void Arm64Mir2Lir::GenFillArrayData(MIR* mir, DexOffset table_offset, RegLocation rl_src) {
- const uint16_t* table = mir_graph_->GetTable(mir, table_offset);
- // Add the table to the list - we'll process it later
- FillArrayData *tab_rec =
- static_cast<FillArrayData*>(arena_->Alloc(sizeof(FillArrayData), kArenaAllocData));
- tab_rec->table = table;
- tab_rec->vaddr = current_dalvik_offset_;
- uint16_t width = tab_rec->table[1];
- uint32_t size = tab_rec->table[2] | ((static_cast<uint32_t>(tab_rec->table[3])) << 16);
- tab_rec->size = (size * width) + 8;
-
- fill_array_data_.push_back(tab_rec);
-
- // Making a call - use explicit registers
- FlushAllRegs(); /* Everything to home location */
- LoadValueDirectFixed(rl_src, rs_x0);
- LoadWordDisp(rs_xSELF, QUICK_ENTRYPOINT_OFFSET(8, pHandleFillArrayData).Int32Value(),
- rs_xLR);
- // Materialize a pointer to the fill data image
- NewLIR3(kA64Adr2xd, rx1, 0, WrapPointer(tab_rec));
- ClobberCallerSave();
- LIR* call_inst = OpReg(kOpBlx, rs_xLR);
- MarkSafepointPC(call_inst);
-}
-
-/*
* Handle unlocked -> thin locked transition inline or else call out to quick entrypoint. For more
* details see monitor.cc.
*/
diff --git a/compiler/dex/quick/arm64/codegen_arm64.h b/compiler/dex/quick/arm64/codegen_arm64.h
index 93d9b34..510bd4c 100644
--- a/compiler/dex/quick/arm64/codegen_arm64.h
+++ b/compiler/dex/quick/arm64/codegen_arm64.h
@@ -183,7 +183,6 @@
void GenEntrySequence(RegLocation* ArgLocs, RegLocation rl_method) OVERRIDE;
void GenExitSequence() OVERRIDE;
void GenSpecialExitSequence() OVERRIDE;
- void GenFillArrayData(MIR* mir, DexOffset table_offset, RegLocation rl_src) OVERRIDE;
void GenFusedFPCmpBranch(BasicBlock* bb, MIR* mir, bool gt_bias, bool is_double) OVERRIDE;
void GenFusedLongCmpBranch(BasicBlock* bb, MIR* mir) OVERRIDE;
void GenSelect(BasicBlock* bb, MIR* mir) OVERRIDE;
diff --git a/compiler/dex/quick/gen_common.cc b/compiler/dex/quick/gen_common.cc
index 3f7ecfe..3a3821f 100644
--- a/compiler/dex/quick/gen_common.cc
+++ b/compiler/dex/quick/gen_common.cc
@@ -496,6 +496,27 @@
}
}
+/*
+ * Array data table format:
+ * ushort ident = 0x0300 magic value
+ * ushort width width of each element in the table
+ * uint size number of elements in the table
+ * ubyte data[size*width] table of data values (may contain a single-byte
+ * padding at the end)
+ *
+ * Total size is 4+(width * size + 1)/2 16-bit code units.
+ */
+void Mir2Lir::GenFillArrayData(MIR* mir, DexOffset table_offset, RegLocation rl_src) {
+ if (kIsDebugBuild) {
+ const uint16_t* table = mir_graph_->GetTable(mir, table_offset);
+ const Instruction::ArrayDataPayload* payload =
+ reinterpret_cast<const Instruction::ArrayDataPayload*>(table);
+ CHECK_EQ(payload->ident, static_cast<uint16_t>(Instruction::kArrayDataSignature));
+ }
+ uint32_t table_offset_from_start = mir->offset + static_cast<int32_t>(table_offset);
+ CallRuntimeHelperImmRegLocation(kQuickHandleFillArrayData, table_offset_from_start, rl_src, true);
+}
+
//
// Slow path to ensure a class is initialized for sget/sput.
//
diff --git a/compiler/dex/quick/mips/call_mips.cc b/compiler/dex/quick/mips/call_mips.cc
index 6536c41..8b5bc45 100644
--- a/compiler/dex/quick/mips/call_mips.cc
+++ b/compiler/dex/quick/mips/call_mips.cc
@@ -210,54 +210,6 @@
branch_over->target = target;
}
-/*
- * Array data table format:
- * ushort ident = 0x0300 magic value
- * ushort width width of each element in the table
- * uint size number of elements in the table
- * ubyte data[size*width] table of data values (may contain a single-byte
- * padding at the end)
- *
- * Total size is 4+(width * size + 1)/2 16-bit code units.
- */
-void MipsMir2Lir::GenFillArrayData(MIR* mir, DexOffset table_offset, RegLocation rl_src) {
- const uint16_t* table = mir_graph_->GetTable(mir, table_offset);
- // Add the table to the list - we'll process it later
- FillArrayData* tab_rec =
- reinterpret_cast<FillArrayData*>(arena_->Alloc(sizeof(FillArrayData),
- kArenaAllocData));
- tab_rec->table = table;
- tab_rec->vaddr = current_dalvik_offset_;
- uint16_t width = tab_rec->table[1];
- uint32_t size = tab_rec->table[2] | ((static_cast<uint32_t>(tab_rec->table[3])) << 16);
- tab_rec->size = (size * width) + 8;
-
- fill_array_data_.push_back(tab_rec);
-
- // Making a call - use explicit registers
- FlushAllRegs(); /* Everything to home location */
- LockCallTemps();
- LoadValueDirectFixed(rl_src, rs_rMIPS_ARG0);
-
- // Must prevent code motion for the curr pc pair
- GenBarrier();
- NewLIR0(kMipsCurrPC); // Really a jal to .+8
- // Now, fill the branch delay slot with the helper load
- RegStorage r_tgt = LoadHelper(kQuickHandleFillArrayData);
- GenBarrier(); // Scheduling barrier
-
- // Construct BaseLabel and set up table base register
- LIR* base_label = NewLIR0(kPseudoTargetLabel);
-
- // Materialize a pointer to the fill data image
- NewLIR4(kMipsDelta, rMIPS_ARG1, 0, WrapPointer(base_label), WrapPointer(tab_rec));
-
- // And go...
- ClobberCallerSave();
- LIR* call_inst = OpReg(kOpBlx, r_tgt); // ( array*, fill_data* )
- MarkSafepointPC(call_inst);
-}
-
void MipsMir2Lir::GenMoveException(RegLocation rl_dest) {
int ex_offset = Thread::ExceptionOffset<4>().Int32Value();
RegLocation rl_result = EvalLoc(rl_dest, kRefReg, true);
diff --git a/compiler/dex/quick/mips/codegen_mips.h b/compiler/dex/quick/mips/codegen_mips.h
index be94019..bd709f3 100644
--- a/compiler/dex/quick/mips/codegen_mips.h
+++ b/compiler/dex/quick/mips/codegen_mips.h
@@ -116,7 +116,6 @@
void GenEntrySequence(RegLocation* ArgLocs, RegLocation rl_method);
void GenExitSequence();
void GenSpecialExitSequence();
- void GenFillArrayData(MIR* mir, DexOffset table_offset, RegLocation rl_src);
void GenFusedFPCmpBranch(BasicBlock* bb, MIR* mir, bool gt_bias, bool is_double);
void GenFusedLongCmpBranch(BasicBlock* bb, MIR* mir);
void GenSelect(BasicBlock* bb, MIR* mir);
diff --git a/compiler/dex/quick/mir_to_lir.h b/compiler/dex/quick/mir_to_lir.h
index 3de4c56..858fb1c 100644
--- a/compiler/dex/quick/mir_to_lir.h
+++ b/compiler/dex/quick/mir_to_lir.h
@@ -832,6 +832,7 @@
void GenNewArray(uint32_t type_idx, RegLocation rl_dest,
RegLocation rl_src);
void GenFilledNewArray(CallInfo* info);
+ void GenFillArrayData(MIR* mir, DexOffset table_offset, RegLocation rl_src);
void GenSput(MIR* mir, RegLocation rl_src, OpSize size);
// Get entrypoints are specific for types, size alone is not sufficient to safely infer
// entrypoint.
@@ -1321,7 +1322,6 @@
virtual void GenEntrySequence(RegLocation* ArgLocs, RegLocation rl_method) = 0;
virtual void GenExitSequence() = 0;
- virtual void GenFillArrayData(MIR* mir, DexOffset table_offset, RegLocation rl_src) = 0;
virtual void GenFusedFPCmpBranch(BasicBlock* bb, MIR* mir, bool gt_bias, bool is_double) = 0;
virtual void GenFusedLongCmpBranch(BasicBlock* bb, MIR* mir) = 0;
diff --git a/compiler/dex/quick/x86/call_x86.cc b/compiler/dex/quick/x86/call_x86.cc
index 441ec9e..86efc1e 100644
--- a/compiler/dex/quick/x86/call_x86.cc
+++ b/compiler/dex/quick/x86/call_x86.cc
@@ -126,54 +126,6 @@
branch_over->target = target;
}
-/*
- * Array data table format:
- * ushort ident = 0x0300 magic value
- * ushort width width of each element in the table
- * uint size number of elements in the table
- * ubyte data[size*width] table of data values (may contain a single-byte
- * padding at the end)
- *
- * Total size is 4+(width * size + 1)/2 16-bit code units.
- */
-void X86Mir2Lir::GenFillArrayData(MIR* mir, DexOffset table_offset, RegLocation rl_src) {
- const uint16_t* table = mir_graph_->GetTable(mir, table_offset);
- // Add the table to the list - we'll process it later
- FillArrayData* tab_rec =
- static_cast<FillArrayData*>(arena_->Alloc(sizeof(FillArrayData), kArenaAllocData));
- tab_rec->table = table;
- tab_rec->vaddr = current_dalvik_offset_;
- uint16_t width = tab_rec->table[1];
- uint32_t size = tab_rec->table[2] | ((static_cast<uint32_t>(tab_rec->table[3])) << 16);
- tab_rec->size = (size * width) + 8;
-
- fill_array_data_.push_back(tab_rec);
-
- // Making a call - use explicit registers
- FlushAllRegs(); /* Everything to home location */
- RegStorage array_ptr = TargetReg(kArg0, kRef);
- RegStorage payload = TargetPtrReg(kArg1);
- RegStorage method_start = TargetPtrReg(kArg2);
-
- LoadValueDirectFixed(rl_src, array_ptr);
- // Materialize a pointer to the fill data image
- if (base_of_code_ != nullptr) {
- // We can use the saved value.
- RegLocation rl_method = mir_graph_->GetRegLocation(base_of_code_->s_reg_low);
- if (rl_method.wide) {
- LoadValueDirectWide(rl_method, method_start);
- } else {
- LoadValueDirect(rl_method, method_start);
- }
- store_method_addr_used_ = true;
- } else {
- NewLIR1(kX86StartOfMethod, method_start.GetReg());
- }
- NewLIR2(kX86PcRelAdr, payload.GetReg(), WrapPointer(tab_rec));
- OpRegReg(kOpAdd, payload, method_start);
- CallRuntimeHelperRegReg(kQuickHandleFillArrayData, array_ptr, payload, true);
-}
-
void X86Mir2Lir::GenMoveException(RegLocation rl_dest) {
int ex_offset = cu_->target64 ?
Thread::ExceptionOffset<8>().Int32Value() :
diff --git a/compiler/dex/quick/x86/codegen_x86.h b/compiler/dex/quick/x86/codegen_x86.h
index 8edfc01..b3544da 100644
--- a/compiler/dex/quick/x86/codegen_x86.h
+++ b/compiler/dex/quick/x86/codegen_x86.h
@@ -245,7 +245,6 @@
void GenEntrySequence(RegLocation* ArgLocs, RegLocation rl_method) OVERRIDE;
void GenExitSequence() OVERRIDE;
void GenSpecialExitSequence() OVERRIDE;
- void GenFillArrayData(MIR* mir, DexOffset table_offset, RegLocation rl_src) OVERRIDE;
void GenFusedFPCmpBranch(BasicBlock* bb, MIR* mir, bool gt_bias, bool is_double) OVERRIDE;
void GenFusedLongCmpBranch(BasicBlock* bb, MIR* mir) OVERRIDE;
void GenSelect(BasicBlock* bb, MIR* mir) OVERRIDE;